Bio-inspired motor primitives for controlling leg exoskeletons

Laburpena

Locomotion assistive devices have been at the focus of many research projects for the recent past years. The reason for this increasing interest arises from three main areas: a growing sector of elderly population that needs and wants to continue being active in society, the fast development of new technologies allowing young disabled patients to enhance their locomotion skills, and the industrial and military sectors using exoskeletons as power augmentation devices. In order to improve the design and control of such assistive devices, an emerging approach deals with bio-inspiration, i.e. capturing biological principles by reproducing desirable features of human locomotion in the devices themselves. In this dissertation, a novel assistive controller based on the bio-inspired concept of motor primitives is presented. Motor primitives are considered as fundamental units of action, whose neural coding is located in the spinal cord. Through proper recombination, these primitives can generate a high-dimensional set of stimulations activating the body muscles and thus generating movement. In this dissertation, the particular use of motor primitives for assisting different rhythmic locomotion modes (i.e., ground-level walking at several cadences, and ascending and descending stairs) is explored. In order to emulate the biological process within the presented controller, artificial motor primitives are generated from locomotion data of the literature reporting the different tasks. Then, these primitives are combined with the appropriate weights to produce muscle stimulations. Finally, these stimulations enter a virtual musculoskeletal model, which produces reference torque profiles. Additionally, other biologically-inspired locomotion mechanisms based on feedback were explored: short-loop reflexes and torso-balance control. The combination of the feedback mechanisms together with the artificial primitives was validated through different simulations. First, the controller was validated offline by proving the suitability of the torque outputs when trying different combinations and using real human recorded data as kinematic inputs. Second, the artificial primitive concept was tested in a different scenario. An autonomous bipedal walking model was developed and connected to the artificial primitives and reflexes. From this second approach, simulated bipedal walking was achieved by combining these intrinsically different bio-inspired sources of muscular stimulation. This simulation tool provides the ideal environment to further explore the relative contributions of these mechanisms in human locomotion. In the following chapters, this dissertation focused on the actual validation of the controller using artificial primitives through three different sets of experiments with human participants. These experiments particularly focused on the capacity of the controller to provide assistance for different locomotion modes, both with healthy and disabled subjects. The first set of experiments served as a primary validation of the controller to work in real-time and provide assistance. These trials were performed with healthy subjects during ground-level walking while assistance was provided at the hip level with an active pelvis module. These experimental activities highlighted the capacity of volunteers to naturally interact with the device and benefit from the assistance regarding their physical effort. The second set of experiments aimed at testing the performance of the controller when being used with gait-impaired subjects. One transfemoral amputee was recruited and performed two sets of trials: one in treadmill walking and another in ground-level walking, both while being assisted at the hip level. This experiment emphasized the capacity of the controller to adapt to mild-impaired gaits and to interact with the amputee. In this case, some improvements were observed on the symmetry of the range of movement at the level of the hip, and the subject reported to feel assisted. However, metabolic indicators were not clear enough to draw any conclusion regarding metabolic effort. Finally, the last set of experiments aimed at proving the full capability of the controller by assisting healthy subjects during all possible programmed locomotion tasks, i.e., walking, stair ascending, and stair descending. Importantly, subjects performed all the tasks continuously, therefore testing also several transitions between them. These experiments highlighted the capacity of the controller to provide relevant assistive torques and to effectively handle transitions between the tasks. Subjects displayed a natural interaction with the device. Moreover, they significantly decreased the time needed to complete the track when the assistance was provided, as compared to wearing the device with no assistance.